Continental-scale distributions of dust-associatedbacteria and fungiAlbert Barberána, Joshua Ladaub, Jonathan W. Leffa,c, Katherine S. Pollardb, Holly L. Menningerd, Robert R. Dunnd,e,and Noah Fierera,c,1

aCooperative Institute for Research in Environmental Sciences, University of Colorado, Boulder, CO 80309; bGladstone Institutes, University of California,San Francisco, CA 94158; cDepartment of Ecology and Evolutionary Biology, University of Colorado, Boulder, CO 80309; dDepartment of Biological Sciences,North Carolina State University, Raleigh, NC 27695; and eCenter for Macroecology, Evolution and Climate, Natural History Museum of Denmark, Universityof Copenhagen, Copenhagen DK-2100, Denmark

Edited by William H. Schlesinger, Cary Institute of Ecosystem Studies, Millbrook, NY, and approved March 20, 2015 (received for review October 31, 2014)

It has been known for centuries that microorganisms are ubiqui-tous in the atmosphere, where they are capable of long-distancedispersal. Likewise, it is well-established that these airbornebacteria and fungi can have myriad effects on human health, aswell as the health of plants and livestock. However, we have alimited understanding of how these airborne communities varyacross different geographic regions or the factors that structurethe geographic patterns of near-surface microbes across largespatial scales. We collected dust samples from the external surfacesof ∼1,200 households located across the United States to under-stand the continental-scale distributions of bacteria and fungi inthe near-surface atmosphere. The microbial communities werehighly variable in composition across the United States, but thegeographic patterns could be explained by climatic and soil vari-ables, with coastal regions of the United States sharing similarairborne microbial communities. Although people living in moreurbanized areas were not found to be exposed to distinct outdoorair microbial communities compared with those living in more ru-ral areas, our results do suggest that urbanization leads to homog-enization of the airborne microbiota, with more urban communitiesexhibiting less continental-scale geographic variability than morerural areas. These results provide our first insight into the continen-tal-scale distributions of airborne microbes, which is informationthat could be used to identify likely associations between microbialexposures in outdoor air and incidences of disease in crops, live-stock, and humans.

aerobiology | microbial ecology | microbial dispersal | urbanization |allergens

For nearly 2 centuries, we have known that microbes areubiquitous in dust and outdoor air (1–3). In the near-surface

atmosphere, microbial cells likely account for a significant frac-tion of aerosolized organic carbon (4), with microbial cell numberstypically ranging from 104 to 106 cells·m−3 over land (5). Thismeans we inhale thousands of microbial cells every hour spentoutdoors, and the potential effects of these airborne microbes onhuman health and the health of plants and animals are wellrecognized. As just one example, there are currently ∼16 millionpeople living in the United States suffering from allergic asthma(6), and there has been considerable attention focused on un-derstanding the airborne and dust-associated microbes influ-encing asthma and how those microbial triggers vary across spaceand time (6–9). In addition, the number of virulent fungal in-fections affecting human populations, wildlife, and plant crops isincreasing (10, 11). The effects of those microbes capable ofatmospheric transport can even extend to entire ecosystems:Microbes in Saharan dust clouds, for example, have been shownto affect the ecology of alpine lakes in Spain (12) and coral reefsin the Caribbean Sea (13).Despite the well-recognized importance of airborne microbes

and the long history of research on microbial transport throughthe atmosphere (3), we have only recently been able to describe

the full extent of microbial diversity found in the atmosphere byusing molecular approaches to characterize the microbial taxathat are difficult to identify via cultivation-dependent ormicroscopy-based surveys. Such molecular approaches not onlyhave yielded new insight into the enormous diversity of airbornemicrobes (14, 15) but also have been instrumental in helping usunderstand how the composition of the airborne microbialcommunities varies across time and space (16, 17). However,nearly all of this work has focused on local-scale variability, ex-amining the bacterial or fungal communities found in outdoor airat selected sites. What is missing is a continental-scale under-standing of microbial diversity in the near-surface atmosphereand its deposition patterns over land. We do not know whetherthere are distinct microbial taxa found in the outdoor air fromdifferent geographic regions, nor do we know what biotic andabiotic factors may be driving the geographic patterns of dust-associated microbial communities across larger spatial scales.There is some evidence to suggest there are regional differencesin exposures to specific bacterial and fungal taxa (18) that may beassociated with geographic patterns in allergenic disease (19),but the continental-scale biogeographic patterns exhibited by thebroad range of microbes that can be found in outdoor airremain undetermined.Here we used dust samples collected from the external sur-

faces of ∼1,200 homes located across the continental UnitedStates to gain our first insight into the continental-scale patternsof bacterial and fungal diversity in the near-surface atmosphere

Significance

We inhale thousands of microbial cells when we breathe inoutdoor air, and some of these airborne microbes can serve aspathogens or triggers of allergic disorders. Using settled dustsamples from ∼1,200 locations, we generated the first atlas, toour knowledge, of airborne bacterial and fungal distributionsacross the continental United States. We found that airbornemicrobial communities, such as terrestrial plants and animals,exhibit nonrandom geographic patterns, and we identified thefactors that shape the continental-scale distributions of microbialtaxa. Furthermore, we found that the airborne microbes foundin urban and more rural areas are not distinct in composition,but the dust-associated communities found in more urbanizedareas are more homogeneous across the United States.

Author contributions: H.L.M., R.R.D., and N.F. designed research; A.B., J.L., and J.W.L.analyzed data; and A.B., K.S.P., R.R.D., and N.F. wrote the paper.

The authors declare no conflict of interest.

This article is a PNAS Direct Submission.

Data deposition: The data reported in this paper have been deposited in the Figsharedata repository (figshare.com/articles/1000homes/1270900) and in the National Center forBiotechnology Information database (BioProject ID PRJNA280363).1To whom correspondence should be addressed. Email: noah.fierer@colorado.edu.

This article contains supporting information online at www.pnas.org/lookup/suppl/doi:10.1073/pnas.1420815112/-/DCSupplemental.

5756–5761 | PNAS | May 5, 2015 | vol. 112 | no. 18 www.pnas.org/cgi/doi/10.1073/pnas.1420815112

Dow

nloa

ded

by g

uest

on

May

20,

202

0

and the factors driving the distributions of these airborne mi-crobial taxa. Specifically, we investigated two questions: First, towhat extent are the diversity and types of bacteria and fungifound in settled dust collected from outside homes a function ofgeographic location, climatic variables, soil characteristics, cropproduction, and land use type? Second, we sought to determinewhether urbanization has a significant effect on bacterial andfungal exposures, given that it has long been assumed that air-borne microbial exposures differ across urban and rural areas(3), with recent work suggesting that these differential microbialexposures may be one explanation for the higher rates of allergicdisorders often observed in more urbanized areas (7, 8).

Results and DiscussionMicrobial Diversity of Dust Samples.Our results stand in contrast toculture-based surveys, which typically show that only a handfulof bacterial or fungal taxa dominate most air and settled dustsamples (e.g., refs. 20, 21). The average dust sample harbored∼4,700 and ∼1,400 bacterial and fungal phylotypes, respectively,of the total of ∼112,000 and ∼57,000 bacterial and fungal phy-lotypes observed across all samples (SI Appendix, Fig. S3). Thenumber of different bacterial or fungal phylotypes (richness)found in each sample (both correlated; Pearson’s r = 0.69; P <0.001) did not vary in any consistent manner with geographicdistance [rM = 0.03 (P = 0.03); rM = 0.02 (P = 0.03); Mantel testsfor bacterial and fungal richness differences, respectively]. Al-though previous culture-independent studies have shown highlevels of bacterial diversity in dust samples comparable to whatwas observed here (12, 14, 22), we also found an equally im-pressive amount of fungal diversity in the collected dust samples,with 74% of these fungal taxa being undescribed (i.e., not rep-resented in the reference sequence database).Dust samples were dominated by Proteobacteria of the class

Alphaproteobacteria (SI Appendix, Fig. S4) and Ascomycotafrom the orders Capnodiales and Pleosporales (SI Appendix, Fig.S5). Most phylotypes were restricted to a small subset of samples;88% and 94% of the bacterial and fungal phylotypes, respectively,were detected in only 10% of the collected samples. Only 0.03%

and 0.02% of the bacterial and fungal phylotypes were foundin more than 90% of the samples, and these ubiquitous andabundant phylotypes belonged to the genera Sphingomonas andHymenobacter for bacteria and toCladosporium, Toxicocladosporium,and Alternaria for fungi (SI Appendix, Fig. S6); these are taxathat are commonly found in the atmosphere (12, 14, 15, 21). Inother words, although most people are likely exposed to thesecommon taxa when breathing outdoor air, the aerosolized bac-teria and fungi are highly variable in composition across thecontinental United States (see variability in SI Appendix, Figs. S4and S5). Given that many bacteria and fungi are capable of long-distance dispersal through the atmosphere (5, 10, 13), one mightexpect greater homogeneity in the composition of the bacterial andfungal communities found in outdoor air. This is clearly not the case.

Factors Structuring Community Similarity Patterns. The bacterialand fungal communities were highly variable in compositionacross the collected samples (SI Appendix, Figs. S2 and S3), andthis variability was predictable. Community similarity patternsbetween bacteria and fungi were correlated (rM = 0.52; P < 0.01;Mantel test), suggesting similar factors structure these commu-nities in the dust settled on the outside surface of homes acrossthe United States. For both groups of microorganisms, com-munity similarity was related to the geographic location of thesample (in contrast to the pattern for richness), although thisrelationship was weaker for bacteria. Sites that were locatedcloser in proximity tended to share more similar microbialcommunities than sites located further apart [rM = 0.13 (P <0.01); rM = 0.29 (P < 0.01); Mantel test for bacterial and fungalcommunities, respectively]. The effects of geographic distancewere most obvious at local to regional scales. At sites >750 kmapart, there was no significant geographic structure in communitysimilarity (SI Appendix, Fig. S7 and Fig. 1), as evidenced by thesamples collected from coastal regions in the eastern and westernUnited States sharing similar communities (Fig. 1). Althoughmarine bacteria are capable of being aerosolized (23), this bi-coastal pattern is not a result of the presence of marine-associatedmicrobes, as their geographic distribution was patchy, with such

−0.25

0.00

0.25

0.50

5 6 7 8Soil pH

Bac

teri

al N

MD

S

−0.25

0.00

0.25

0.50

50 100 150 200

Mean Annual Precipitation (mm)

Fu

ng

al N

MD

S

A

B

Fig. 1. Maps of overall community similarity (nonmetric multidimensional scaling axis scores) and the relationships between community similarity and envi-ronmental parameters for (A) bacterial communities (r = −0.40; P < 0.001; soil pH) and (B) fungal communities (r = 0.56; P < 0.001; mean annual precipitation).Similar colors indicate bacterial or fungal communities that are more similar in overall community composition. Points represent sampling locations.

Barberán et al. PNAS | May 5, 2015 | vol. 112 | no. 18 | 5757

ENVIRONMEN

TAL

SCIENCE

S

Dow

nloa

ded

by g

uest

on

May

20,

202

0

microbes being rare and only present in a handful of samples (SIAppendix, Fig. S8 and Table S2). As some of these marine bac-terial taxa are also common in many freshwater environments(e.g., Prochlorococcus, Synechococcus), their presence in locationsfar from oceans also may be a result of aerosolization of bacteriafrom lakes or other nearby water bodies.We then determined what factors are correlated with the ob-

served similarity patterns in community composition (Fig. 1).Previous regional-scale studies have identified vegetation, land-use characteristics (17, 24), and climatic conditions (14) as fac-tors influencing airborne microbial communities. Others factorssuch as human population and livestock density or house char-acteristics were included as additional candidate explanatoryvariables in this study. A list of these variables is provided in SIAppendix, Table S1. Our results, using multiple regression ondistance matrices (MRM), indicated that the composition of thedust-associated microbial communities was significantly corre-lated with a combination of soil and climatic variables. The bestpredictors of bacterial community composition, in order of im-portance, were soil pH, mean annual precipitation, net primaryproductivity, and mean annual temperature [overall correlationwith community similarity: rM = 0.25 (P < 0.01), Mantel test;MRM: R2 = 0.06 (P < 0.001); see Fig. 1A for the relationshipbetween the main bacterial ordination axis and soil pH]. Simi-larly, the best predictors of fungal community composition were,in order of importance, mean annual precipitation, soil pH, netprimary productivity, the diversity of vascular plants, and meanannual temperature [overall correlation with community simi-larity: rM = 0.38 (P < 0.01), Mantel test; MRM: R2 = 0.15 (P <0.001); see Fig. 1B for the relationship between the main fungalordination axis and mean annual precipitation]. As climate, soilcharacteristics, and plant productivity are tightly linked, wecannot assign unequivocal cause and effects relationships. That

is, we do not know whether the observed patterns in fungalcommunity structure are driven directly by climate or by differ-ences in plant productivity (which will be approximately collinearwith climate at the continental-scale of inquiry). The same caveatapplies to the apparent effects of soil characteristics and climateon bacterial community structure.The environmental associations were also evident when we

looked at the geographic distributions of individual taxa. Forexample, the strong influence of soil characteristics in shapingdust bacterial community composition is consistent with thedistributions of Cellulomonas, one of the more abundant acti-nobacterial genera (Fig. 2A; Pearson’s correlation with soil pH:r = 0.47; P < 0.001), and Terriglobus, the most abundant genusof Acidobacteria (Fig. 2B; Pearson’s correlation with soil pH:r = −0.31; P < 0.001). These patterns agree with previous workshowing that soil bacterial communities are strongly structuredby pH (25), with Acidobacteria dominating in low-pH soils andactinobacterial taxa being relatively more abundant in high-pHsoils such as those found in deserts (26, 27). Many fungal taxaalso exhibited distinct geographic patterns. Most notably, thegenus Alternaria, which contains species that are well-knowntriggers of allergenic disease (28) and is commonly found in bothindoor and outdoor dust samples (15), was highest in abundancein samples collected from the Great Plains region of the UnitedStates, with relative abundances that often exceeded 50% of thefungal sequences from individual samples (Fig. 2C). Likewise, wefound that the genus Cladosporium, the most abundant genus inour samples (often representing >75% of all fungal sequencesfrom individual samples) and one of the more common fungalgroups identified in other studies of airborne and dust-associatedfungi (e.g., refs. 21, 29), was dominant in humid areas with low-pH soils (Fig. 2D). This biogeographic pattern is noteworthy,given that Cladosporium is a common allergen and members of

0.2

0.4

0.6

Proportion ofAlternaria

0.250.500.75

Proportion ofCladosporium

0.0050.0100.015

Proportion ofCellulomonas

0.0050.0100.0150.020

Proportion ofTerriglobus

A

C

B

D

Fig. 2. Maps showing the relative abundance of specific bacterial and fungal genera across the continental United States: (A) Cellulomonas, (B) Terriglobus,(C) Alternaria, and (D) Cladosporium. Warmer colors indicate higher relative abundances. Points represent sampling locations.

5758 | www.pnas.org/cgi/doi/10.1073/pnas.1420815112 Barberán et al.

Dow

nloa

ded

by g

uest

on

May

20,

202

0

this genus include major plant pathogens. Together, these resultshighlight that many microbial taxa exhibit local- and regional-scalegeographic distribution patterns and that we can build range mapsof bacterial and fungal taxa comparable to the maps long built byplant and animal ecologists to quantify biogeographic patterns andpredict responses to climate change (30).Our results indicate that the microbial communities found in

outdoor air exhibit local and regional-scale geographic patterns,even though airborne microbes are known to disperse manythousands of kilometers (10, 12, 13, 31) and we might expect thecomplexities of wind patterns in the near-surface atmosphere toobscure the biogeographic distributions of microbes in settleddust. Some of the geographic patterns in airborne microbialcommunities could be driven by differences in dispersal capa-bilities. Not all taxa are equally adept at dispersal and survival onthe sampled house surfaces, and this variation in the abilities ofmicrobes to disperse to a given location and persist at that lo-cation could be an important mechanism generating the bio-geographic patterns observed here (32). Likewise, it is importantto mention that many local-scale factors that were not effectivelycaptured with our data or statistical models could be responsiblefor generating some of the observed biogeographic patterns. Forexample, the types of vegetation found neighboring the homes orother local-scale sources of microbes to the atmosphere (e.g., con-struction sites, dirt roads) may be influencing the structure of thesedust-associated communities at smaller spatial scales.

Effect of Urbanization. Urbanization tends to homogenize plantand animal communities, with more urbanized areas tending toshare a large number of plant and animal taxa in common (33).At the same time, urbanization promotes the establishment ofnonnative plant and animal species, and thus, cities can oftenhave higher species richness than surrounding rural areas (e.g., refs.34, 35). We do not know how urbanization affects airborne microbialcommunities, with some studies suggesting microbial diversity shouldbe lower in more urbanized areas (8) and other studies showing littleto no effect of urbanization on microbial diversity (17). This questionis important, given that rates of allergenic asthma are higher for thechildren living in more urbanized areas (reviewed in ref. 9). Althoughmany factors contribute to allergenic disease (19), it has been hy-pothesized that these geographic differences in allergy rates can beattributed to people living in more urbanized areas being exposed tolower levels of microbial diversity (7, 8). However, at least outdoors,we found no significant effect of urbanization on bacterial or fungaldiversity levels (P > 0.05, Mann-Whitney test; SI Appendix, Fig. S9).Likewise, we found that more urbanized sites did not have mi-crobial communities compositionally distinct from those found inless-urbanized or rural locations (R2 < 0.005 for both bacterialand fungal communities, permutational analysis of variance(PERMANOVA); SI Appendix, Fig. S9). We also did not ob-serve any effect of urbanization on bacterial or fungal communitycomposition when we treated urbanization as a quantitativevariable (i.e., population density, R2 < 0.005, PERMANOVA),rather than a qualitative one (i.e., urban and rural classification).Although we observed no significant effects of urbanization on

the overall diversity and composition of dust-associated microbialcommunities, we expected the specific sources of airborne mi-crobes to vary across urban to rural gradients (24). To test thishypothesis, we identified specific bacterial taxa indicative of spe-cific source environments from which dust-associated microbesare most likely to originate (SI Appendix, Table S2), including soil,marine waters, feces and skin of humans or other animals, orplants. On the basis of these source assignments, we could thentest how urbanization affects the relative importance of specificsources of bacteria to the atmosphere. For example, we wouldexpect fecal bacteria to be more abundant in rural areas close tohigh densities of livestock (24, 36), and skin-associated microbesto be more common in cities and populated areas. However, we

did not detect any strong or significant effect of urbanization onthe proportion of different bacterial indicator taxa (no differencegreater than 1% or P < 0.05, Mann-Whitney test; Fig. 3).Overall, bacterial communities in the dust samples were pri-marily derived from bacteria likely to be found on plants and soil,regardless of the degree of urbanization (Fig. 3), a finding in linewith other work showing that these sources are the dominantcontributors of bacteria to the near-surface atmosphere in mostterrestrial environments (13, 15). The imprint of human-associ-ated bacteria was low and primarily a result of skin, rather thanstool (with the exception of a single rural sample that wasdominated by stool indicator taxa, mainly Bacteroides and Para-bacteroides; Fig. 3). Neither marine waters (and potentially otherwater bodies including lakes or rivers) nor insects appeared to beimportant sources of bacteria of the atmosphere (Fig. 3). In thecase of the few samples (particularly from urban areas) in whichwe observed high abundances of bacteria indicative of insects(notably Rickettsiella, Entomoplasmatales, Buchnera, and Wolba-chia), these extreme values might be a result of the presence ofdead insects (or fragments/depositions of insects) on the outdoorframe of those homes. Although we can demonstrate the impor-tance of plants and soils as sources of microbes in settled dust, ourdata suggest that people living in more rural areas are not exposedoutdoors to bacteria derived from different source environmentsthan those living in more urbanized areas.However, in line with the patterns commonly observed for

plant and animal communities (33), the bacterial and fungalcommunities found in the dust from more urbanized locationstended to be more homogeneous. In other words, the microbialcommunities found in cities were not as spatially variable acrossthe continental United States as in those communities found inmore rural areas (Fig. 4). People living in more urbanized areasare not likely exposed to different airborne bacteria and fungioutdoors than those living in more rural areas, but urban areastend to harbor microbial communities that are less spatially

Chloroplasts Soil indicator taxa Skin indicator taxa

Stool indicator taxa Insect indicator taxa Sea indicator taxa

0.01

0.10

0.25

0.50

0.01

0.10

0.25

0.50

Rural Urban Rural Urban Rural Urban

Pro

po

rtio

n

Fig. 3. Proportion of bacterial sequences identified as indicator taxa ofpotential source environments in urban and rural samples. Chloroplasts wereused as indicators of plant influence. Source environments are ordered fromleft to right and top to bottom according to their median value (medians areshown as horizontal lines). Note that because of the highly skewed distri-bution, the y axis is squared.

Barberán et al. PNAS | May 5, 2015 | vol. 112 | no. 18 | 5759

ENVIRONMEN

TAL

SCIENCE

S

Dow

nloa

ded

by g

uest

on

May

20,

202

0

variable than those found in more rural areas. This observedhomogenization of dust-associated microbial communities mightbe a result of a homogenization of environmental conditions andland-use types in urban areas (33) or the homogenization inplant communities or other potential biotic sources of bacteriaand fungi to the atmosphere.

MethodsSample Collection. Outdoor dust samples were collected by volunteers par-ticipating in the Wild Life of Our Homes project (homes.yourwildlife.org), acontinental-scale citizen science project. We recruited participants from all50 states and the District of Columbia through our website, social media,and email campaigns over the period from January 2012 to March 2013.Enrolled participants (n = 1,430) were provided a written informed consentform approved by the North Carolina State University’s Human ResearchCommittee (approval no. 2177), as well as instructions for sampling and amicrobe sampling kit containing dual-tipped sterile BBL CultureSwabs. Herewe focus on a subset of samples that participants collected from the upperdoor trim on the outside surface of an exterior door, a sampling location thatis found in every home, is unlikely to be cleaned frequently, and serves as apassive collector of outdoor aerosols and dust with little to no direct contactfrom the home occupants. The amount of time dust has accumulated on thesesurfaces is unknown and likely variable across homes, so we cannot use thissample set to assess temporal variation in the microbial communities. Partici-pants returned swabs by first-class mail over the period from March 2012 toMay 2013, and these swabs were stored in a −20 °C freezer until processed.A map of sampling locations and information on the number of samplescollected per US state are provided in SI Appendix, Fig. S1.

Molecular Analyses. Swabs were prepared for sequencing using the direct PCRapproach described previously (37). Swab tips were placed directly into 2-mL96-well plates (Axygen Inc.), along with the appropriate negative controlsamples. Plates were processed using the Extract-N-Amp PCR kit (Sigma-Aldrich, Inc.), following a modified version of the manufacturers’ instructions.After each well received 250 μL of the Extract-N-Amp Extraction solution,the plate was sealed securely with a 96-round well Impermamat SiliconSealing Mat (Axygen, Inc.) and heated at 90 °C for 10 min in a dry bath.Extract-N-Amp Dilution solution was then added to the wells at a 1:1 ratio tothe extraction solution and mixed gently by pipetting. The plate wasresealed with the mat and stored at 4 °C. PCR was conducted in 20-μL trip-licate reactions per sample, using 10 μL Extract-N-Amp Ready Mix, 1 μL of theforward and reverse primers, 5 μL PCR-grade water, and 4 μL of the Extract-N-Amp sample solutions from the 96-well plate.

Microbial diversitywas assessed using high-throughput sequencingmethodsto characterize the variation in marker gene sequences. For bacterial andarchaeal analyses, we sequenced the V4 hypervariable region of the 16S rRNAgene using the 515-F (GTGCCAGCMGCCGCGGTAA) and 806-R (GGA-CTACHVGGGTWTCTAAT) primer pair (26). For the fungal analyses, we se-quenced the first internal transcribed spacer (ITS1) region of the rRNA operon,using the ITS1-F (CTTGGTCATTTAGAGGAAGTAA) and ITS2 (GCTGCGTTCTT-CATCGATGC) primer pair (38). The primers included the appropriate Illuminaadapters, with the reverse primers also having an error-correcting 12-bp

barcode unique to each sample to permit multiplexing of samples. Archaealsequences represented a very small proportion of the recovered sequences(0.13% of the 16S rRNA gene sequences). PCR products from all samples werequantified using the PicoGreen dsDNA assay and pooled together in equi-molar concentrations for sequencing on either an Illumina HiSeq or MiSeqinstrument. All sequencing runs were completed at the University of ColoradoNext Generation Sequencing Facility.

Sequence Processing. The 100-bp sequences were demultiplexed using acustom Python script (https://github.com/leffj/helper-code-for-uparse), withquality filtering and phylotype clustering conducted using the UPARSEpipeline (39). For quality filtering, we used a maximum e-value of 0.5 (in-dicating that on average, a maximum of 0.5 nucleotides were incorrectlyassigned in every sequence). Sequences were also dereplicated, and single-ton sequences were removed before phylotype determinations. Represen-tative sequences from the phylotypes that were not ≥75% similar tosequences contained in either the Greengenes 13_8 database (40) or theUNITE May 2014 database (41) for 16S and ITS rRNA sequences, respectively,were discarded. Raw sequences were then mapped to phylotypes at the 97%similarity threshold. Phylotype taxonomy was determined using the Ribo-somal Database Project classifier with a confidence threshold of 0.5 (42),trained on the respective databases for 16S and ITS rRNA sequence. 16SrRNA sequences resulting from direct PCR reagent contaminants (37) werefiltered from all samples by removing those classified as Mycoplasma,Pseudomonas, and Serratia, which represented a median 59%, 9%, and 4%of sequences across the controls, respectively. In addition, sequences repre-senting any phylotypes present in ≥25% of control samples and those clas-sified as mitochondria or chloroplast were removed (except during analysisof plant presence described here).

To remove potential amplicon sequencing biases, we first removed sam-ples with fewer than 10,000 sequences, and thenwe normalized the sequencecounts using a cumulative-sum scaling, which scales counts by dividing up to apercentile empirically determined (43). Richness and community similarityvalues after normalization were highly correlated, with values obtainedafter rarefying to 10,000 sequences per sample (SI Appendix, Fig. S2). Thetotal number of samples included in downstream analyses was 1,187 forbacteria and 1,289 for fungi.

Sample Characterization. Latitude and longitude coordinates were derivedfrom location information (address), and the coordinates were used to obtaingeoreferenced variables for each household. We compiled a set of 31 de-scriptors: elevation, distance to the coast, climatic variables, plant pro-ductivity, soil factors, other organisms’ (including humans and livestock)variables, and house characteristics (SI Appendix, Table S1). Urban and ruralclassifications for all sampled locations were obtained from the 2010 USCensus Bureau data (www.census.gov). Urban areas are classified as denselydeveloped (>50,000 people) residential and commercial territories; 28% ofthe sampled locations were classified as rural.

Identification of Bacterial Source Indicators. To identify potential taxa in-dicative of specific source environments (i.e., soil, oceans, human skin, andhuman/livestock feces), we used Indicator Value analyses (44), as described inref. 22. We used the proportion of chloroplast sequences in each sample as aproxy for plant presence. To identify those bacteria likely derived from in-sects, we compiled a list of exclusive symbiont and/or insect pathogens fromthe literature (45, 46). For the complete list of bacterial indicator taxa, seeSI Appendix, Table S2.

Statistical Analyses. Microbial community similarity was represented by non-metric multidimensional scaling, using the Bray-Curtis dissimilarity distancemetric after Hellinger standardization (47). Bray-Curtis abundance-based dis-similarity matrices were highly correlated with Jaccard incidence-based matrices[rM = 0.86 (P < 0.001); rM = 0.92 (P < 0.001); Mantel test for bacteria and fungi,respectively]. As some of the descriptors were correlated, we reduced collin-earity before our analyses by identifying highly correlated variables by principalcomponent analyses and by examining the variance inflation factor (48). Thefinal set of factors consisted of 18 variables (SI Appendix, Table S1). To findthe best subset of explanatory variables, we first ran bioenv (49) to maximizethe correlation with community dissimilarity. We then performed MRM toestimate the overall explanatory power of the model (50). To test whethercommunity composition was different between urban and rural areas, weused PERMANOVA based on 1,000 permutations (51).

All multivariate statistical analyses were carried out in the R environment(www.r-project.org), using the vegan (vegan.r-forge.r-project.org/), labdsv(ecology.msu.montana.edu/labdsv/R/), and ecodist (cran.r-project.org/web/

0.6

0.7

0.8

0.9

1.0

0 1000 2000 3000 4000Spatial distance (km)

Bac

teri

al c

om

mu

nit

y d

issi

mila

rity

(B

rart

is)

0.6

0.7

0.8

0.9

1.0

0 1000 2000 3000 4000Spatial distance (km)

Fu

ng

al c

om

mu

nit

y d

issi

mila

rity

(B

rart

is)A B

RuralUrban

RuralUrban

Fig. 4. Relationship between community dissimilarity and spatial distancefor (A) bacterial and (B) fungal urban and rural samples. The slopes weresignificantly different in both cases (bacteria urban = 0.9·10−5 ± 1.1·10−7;bacteria rural = 1.7·10−5 ± 3.9·10−7; fungi urban = 2.6·10−5 ± 1.3·10−7; fungirural = 4.1·10−5 ± 4.4·10−7).

5760 | www.pnas.org/cgi/doi/10.1073/pnas.1420815112 Barberán et al.

Dow

nloa

ded

by g

uest

on

May

20,

202

0

packages/ecodist/) packages. Nonmetric multidimensional scaling axis scoresand the proportional abundances of taxa were mapped by inverse distanceweighting interpolation on 100 × 100 grid cells, using the gstat package(https://r-forge.r-project.org/projects/gstat/).

ConclusionWe show that the continental-scale distributions of bacteria and fungiin settled dust are correlated with regional environmental factors.These geographic patterns are surprisingly robust, considering thepotential for long-distance dispersal of microbes in aerosolized dustparticles, the complexity of wind patterns in the near-surface atmo-sphere, and the difficulties associated with quantifying specific local-scale factors that may influence the types of bacteria and fungi thatcan be introduced into the atmosphere.The health effects of this work remain unknown. We do know

that bacterial and fungal exposures are distinct across the UnitedStates, and some of this variability can be explained by envi-ronmental conditions; namely, climatic and soil characteristics.

We also know that allergen sensitivities and incidences of asthma,including allergen sensitivities to particular microbial triggers,can vary across the United States (28). Our study was notdesigned to collect specific allergen sensitivity data for the in-dividuals living in each home, but clearly there are opportunitiesfor future work investigating how the diversity, distribution, andcomposition of these dust-associated bacteria and fungi mayshape incidences of disease across the United States.

ACKNOWLEDGMENTS. We thank the many volunteers who participatedin the Wild Life of Our Homes study for collecting home dust samples.Funding for this work was provided by a grant from the A.P. SloanMicrobiology of the Built Environment Program (to N.F. and R.R.D.),with additional support from the US National Science Foundation(DEB0953331 to N.F.). A.B. is supported by a James S. McDonnellPostdoctoral Fellowship. Support was also provided by the National ScienceFoundation (DMS-1069303 to K.S.P. and J.W.L.), the Gordon and BettyMoore Foundation (3300), the Gladstone Institutes, and a gift from the SanSimeon Fund.

1. Darwin C (1846) An account of the fine dust which often falls on vessels in the Atlanticocean. Q J Geol Soc 2:26–30.

2. Pasteur L (1861) Mémoire sur les corpuscules organisés qui existent dans l’atmo-sphère: Examen de la doctrine des générations spontanées. Ann Sci Nat 4e série:5–68.

3. Tyndall J (1882) Essays on the Floating-Matter of the Air (D. Appleton and Company,New York).

4. Wiedinmyer C, et al. (2009) The contribution of biological particles to observed partic-ulate organic carbon at a remote high altitude site. Atmos Environ 43(28):4278–4282.

5. Burrows SM, Elbert W, Lawrence MG, Pöschl U (2009) Bacteria in the global atmo-sphere- Part 1: Review and synthesis of literature data for different ecosystems.Atmos Chem Phys Discuss 9:9263–9280.

6. Denning DW, O’Driscoll BR, Hogaboam CM, Bowyer P, Niven RM (2006) The link be-tween fungi and severe asthma: A summary of the evidence. Eur Respir J 27(3):615–626.

7. Ege MJ, et al.; GABRIELA Transregio 22 Study Group (2011) Exposure to environ-mental microorganisms and childhood asthma. N Engl J Med 364(8):701–709.

8. Hanski I, et al. (2012) Environmental biodiversity, human microbiota, and allergy areinterrelated. Proc Natl Acad Sci USA 109(21):8334–8339.

9. von Mutius E, Vercelli D (2010) Farm living: Effects on childhood asthma and allergy.Nat Rev Immunol 10(12):861–868.

10. Brown JKM, Hovmøller MS (2002) Aerial dispersal of pathogens on the global andcontinental scales and its impact on plant disease. Science 297(5581):537–541.

11. Fisher MC, et al. (2012) Emerging fungal threats to animal, plant and ecosystemhealth. Nature 484(7393):186–194.

12. Barberán A, Henley J, Fierer N, Casamayor EO (2014) Structure, inter-annual recurrence,and global-scale connectivity of airborne microbial communities. Sci Total Environ487:187–195.

13. Prospero JM, Blades E, Mathison G, Naidu R (2005) Interhemispheric transport of vi-able fungi and bacteria from Africa to the Caribbean with soil dust. Aerobiologia 21:1–19.

14. Brodie EL, et al. (2007) Urban aerosols harbor diverse and dynamic bacterial pop-ulations. Proc Natl Acad Sci USA 104(1):299–304.

15. Fröhlich-Nowoisky J, Pickersgill DA, Després VR, Pöschl U (2009) High diversity offungi in air particulate matter. Proc Natl Acad Sci USA 106(31):12814–12819.

16. Fierer N, et al. (2008) Short-term temporal variability in airborne bacterial and fungalpopulations. Appl Environ Microbiol 74(1):200–207.

17. Bowers RM, McLetchie S, Knight R, Fierer N (2011) Spatial variability in airbornebacterial communities across land-use types and their relationship to the bacterialcommunities of potential source environments. ISME J 5(4):601–612.

18. Amend AS, Seifert KA, Samson R, Bruns TD (2010) Indoor fungal composition isgeographically patterned and more diverse in temperate zones than in the tropics.Proc Natl Acad Sci USA 107(31):13748–13753.

19. Salo PM, et al. (2008) Exposure to multiple indoor allergens in US homes and its re-lationship to asthma. J Allergy Clin Immunol 121(3):678–684.

20. Tong Y, Lighthart B (1999) Diurnal distribution of total and culturable atmosphericbacteria at a rural site. Aerosol Sci Technol 30(2):246–254.

21. Shelton BG, Kirkland KH, Flanders WD, Morris GK (2002) Profiles of airborne fungi inbuildings and outdoor environments in the United States. Appl Environ Microbiol68(4):1743–1753.

22. Dunn RR, Fierer N, Henley JB, Leff JW, Menninger HL (2013) Home life: Factorsstructuring the bacterial diversity found within and between homes. PLoS ONE 8(5):e64133.

23. Blanchard DC, Syzdek L (1970) Mechanism for the water-to-air transfer and concen-tration of bacteria. Science 170(3958):626–628.

24. Bowers RM, et al. (2011) Sources of bacteria in outdoor air across cities in the mid-western United States. Appl Environ Microbiol 77(18):6350–6356.

25. Lauber CL, Hamady M, Knight R, Fierer N (2009) Pyrosequencing-based assessment ofsoil pH as a predictor of soil bacterial community structure at the continental scale.Appl Environ Microbiol 75(15):5111–5120.

26. Fierer N, et al. (2012) Cross-biome metagenomic analyses of soil microbial commu-nities and their functional attributes. Proc Natl Acad Sci USA 109(52):21390–21395.

27. Lee CK, Barbier BA, Bottos EM, McDonald IR, Cary SC (2012) The Inter-Valley SoilComparative Survey: The ecology of Dry Valley edaphic microbial communities. ISME J6(5):1046–1057.

28. Salo PM, et al. (2006) Exposure to Alternaria alternata in US homes is associated withasthma symptoms. J Allergy Clin Immunol 118(4):892–898.

29. Adams RI, Miletto M, Taylor JW, Bruns TD (2013) The diversity and distribution offungi on residential surfaces. PLoS ONE 8(11):e78866.

30. MacArthur RH (1972) Geographical Ecology: Patterns in the Distribution of Species(Harper & Rowe Publishers, New York).

31. Bovallius A, Bucht B, Roffey R, Anäs P (1978) Long-range air transmission of bacteria.Appl Environ Microbiol 35(6):1231–1232.

32. Peay KG, Schubert MG, Nguyen NH, Bruns TD (2012) Measuring ectomycorrhizalfungal dispersal: Macroecological patterns driven by microscopic propagules. MolEcol 21(16):4122–4136.

33. McKinney ML (2006) Urbanization as a major cause of biotic homogenization. BiolConserv 127(3):247–260.

34. Hope D, et al. (2003) Socioeconomics drive urban plant diversity. Proc Natl Acad SciUSA 100(15):8788–8792.

35. Kühn I, Brandl R, Klotz S (2004) The flora of German cities is naturally species rich. EvolEcol Res 6:749–764.

36. Bakutis B, Monstviliene E, Januskeviciene G (2004) Analyses of airborne contamina-tion with bacteria, endotoxins and dust in livestock barns and poultry houses. ActaVet (Brno) 73(2):283–289.

37. Flores GE, Henley JB, Fierer N (2012) A direct PCR approach to accelerate analyses ofhuman-associated microbial communities. PLoS ONE 7(9):e44563.

38. McGuire KL, et al. (2013) Digging the New York City Skyline: Soil fungal communitiesin green roofs and city parks. PLoS ONE 8(3):e58020.

39. Edgar RC (2013) UPARSE: Highly accurate OTU sequences from microbial ampliconreads. Nat Methods 10(10):996–998.

40. McDonald D, et al. (2012) An improved Greengenes taxonomy with explicit ranks forecological and evolutionary analyses of bacteria and archaea. ISME J 6(3):610–618.

41. Abarenkov K, et al. (2010) The UNITE database for molecular identification of fungi—recent updates and future perspectives. New Phytol 186(2):281–285.

42. Wang Q, Garrity GM, Tiedje JM, Cole JR (2007) Naive Bayesian classifier for rapidassignment of rRNA sequences into the new bacterial taxonomy. Appl Environ Mi-crobiol 73(16):5261–5267.

43. Paulson JN, Stine OC, Bravo HC, Pop M (2013) Differential abundance analysis formicrobial marker-gene surveys. Nat Methods 10(12):1200–1202.

44. Dufrêne M, Legendre P (1997) Species assemblages and indicator species: The needfor a flexible asymmetrical approach. Ecol Monogr 67(3):345–366.

45. Feldhaar H (2011) Bacterial symbionts as mediators of ecologically important traits ofinsect hosts. Ecol Entomol 36(5):533–543.

46. Engel P, Moran NA (2013) The gut microbiota of insects - diversity in structure andfunction. FEMS Microbiol Rev 37(5):699–735.

47. Legendre P, Gallagher E (2001) Ecologically meaningful transformations for ordina-tion of species data. Oecologia 129(2):271–280.

48. Dormann CF, et al. (2013) Collinearity: A review of methods to deal with it and asimulation study evaluating their performance. Ecography 36(1):27–46.

49. Clarke K, Ainsworth M (1993) A method of linking multivariate community structureto environmental variables. Mar Ecol Prog Ser 92:205–219.

50. Lichstein JW (2006) Multiple regression on distance matrices: A multivariate spatialanalysis tool. Plant Ecol 188(2):117–131.

51. Anderson MJ (2001) A new method for non-parametric multivariate analysis ofvariance. Austral Ecol 26(1):32–46.

Barberán et al. PNAS | May 5, 2015 | vol. 112 | no. 18 | 5761

ENVIRONMEN

TAL

SCIENCE

S

Dow

nloa

ded

by g

uest

on

May

20,

202

0